The design system is becoming the brief
Coinbase’s experiment suggests that better context—not just better models—makes AI-assisted implementation more reliable.
average reduction in implementation cost in Coinbase’s Code Connect test
The Story
Coinbase tested its design system with and without Code Connect, giving coding agents clearer access to production components, tokens, and patterns. The result was a measurable improvement in speed, cost, and component selection. ([figma.com](https://www.figma.com/blog/how-coinbase-used-code-connect-to-shrink-token-costs/))
The useful idea here is less about one Figma feature and more about design documentation as operational infrastructure. If an agent has to infer your visual language from screenshots, it will improvise; if it can read the system, it can extend it.
For Studio Wensday, this points toward a practical service: turn brand rules into usable, structured context—tokens, components, examples, exclusions, and voice—so a small team can build more without losing the plot.
Studio Wensday Angle
Treat brand systems as machine-readable creative direction.